The whole machine behind the feed, end to end: how each day's literature is fetched, de-duplicated, and ranked; the evidence signals attached to every paper; how the AI is kept grounded in real articles; and, plainly, how it can (rarely) be wrong. The ranking is auditable, the signals come straight from PubMed, and none of it is a black box. The corpus today holds 72,799 articles across 5,135 journals, and grows daily.
Relevance is computed from free, auditable signals, not a paid black-box per-article LLM pass:
You set the relevance bar; only what clears it reaches your feed. The default sort is newest-first; an optional Impact sort blends citation impact (RCRRelative Citation Ratio: a citation metric from the NIH that is normalized for field and age, so 1.0 is roughly the median NIH-funded paper. It lets you compare impact fairly across specialties and years., see below) with a recency decay, so important recent and older work surfaces without burying brand-new papers. Crucially, relevance is not a quality grade: it's topical centrality. Quality is what the appraisal is for.
Corpus is a retrieval-augmentedThe AI is handed the actual indexed abstracts to answer from, instead of relying on its training memory, so its answers are tied to real, citable papers. assistant grounded in the corpus: it reads the indexed abstracts and, for open-access papers, the full text (the methods and results, fetched from Europe PMCA free European database of life-science articles and open-access full text, the primary source for funding details and full-text mining. and COREA free global aggregator of open-access research from repositories worldwide. Used as a second source of full text for open-access papers that are not in Europe PMC.), points you to the sources, and is instructed to quote the paper's actual numbers and never invent them. It works four ways: ask about one paper, ask across your library, prep a journal club, and explain the statistics in a paper. It reports what the papers say; it does not give medical advice, and the appraisal stays yours.
Beyond the curated feed: full PubMed search for everything else, a per-specialty landmark-trials library, semantic search that ranks your library by meaning, and related papers: embedding nearest-neighbours that turn any article into a jumping-off point. One search bar handles both: plain language searches by meaning, and native PubMed syntax (field tags, AND/OR/NOT) runs as a PubMed query. You can also track any clinical question forever: describe it in plain English and it becomes a tracked topic, a standing search by meaning that surfaces new matches as the library grows each day.
The library and search now default to high-quality evidence: RCTs, meta-analyses, and guidelines, the studies worth discussing, with one click to widen to everything. A separate Guidelines library keeps practice guidelines, organized by specialty, out of the article feed so you can quickly check where a study agrees or differs. Each guideline carries an AI-formatted quick-reference card (scope, population, key recommendations, grading), generated once from the guideline's abstract and labeled as such, so you can skim the gist before opening the source. It is a convenience summary, not the guideline itself, see section 9.
Foreground is built mainly to consume the science, but the regulatory and industry world around it sometimes matters. Radar applies the same focused, filterable approach to that adjacent world: FDA approvals, clearances (including De Novo first-of-kind devices), recalls (with hazard severity), drug shortages, new clinical trials, Medicare coverage decisions (national and local), FDA and society guidance, and healthtech, pulled from free public sources (openFDAThe U.S. Food and Drug Administration's free public API for drug and device approvals, recalls, and shortages., ClinicalTrials.govThe U.S. public registry of clinical trials, a source for newly posted and updated studies., CMS, ex-US regulators including the EMA and MHRA, CDC MMWR, AHA/ACC statements, preprints). Every item is tagged by topic (authoritatively where possible, via FDA review panels, the NLM drug-class vocabulary, and FDA's own AI-enabled-device list, with keyword matching as a fallback), ranked by a transparent rule-based importance score (hazard, event type, approval pathway, cross-source corroboration, recency), de-duplicated across sources, and can cross-link a regulatory item to the related papers already in your library. Topics include a dedicated AI in Medicine area alongside the clinical specialties.
Radar also shows AI-written desk briefings, one per desk (regulatory, companies & pipeline, trials & evidence, digital health), each grouping that desk's new items into themes with a short practice implication. A briefing is generated once per day on the operator's key and shared by everyone (never per-click), and the past week's editions stay readable per desk. It is genuinely AI-generated text, see section 9 on where that can be wrong. The underlying items, tags, and ranking are not AI. Radar is a quiet, optional tool, not the main event.
A club is a shared space: everyone signs into the same feed, saves, group notes, and structured appraisals (PICOA standard way to frame a clinical question: Population, Intervention, Comparison, Outcome., design, bias, verdict, which the AI can draft for you to verify). Organize saves into project folders, schedule sessions around papers with a presenter and agenda, export a folder's citations (BibTeX / RIS / NBIBBibTeX, RIS, and NBIB are standard citation file formats that reference managers (Zotero, EndNote, Mendeley) can import directly. / formatted AMA), and share a folder as a public read-only link.
Today Foreground is open to use at no charge. The daily surveillance and ranking run with no LLM and no key; the AI pieces (the librarian, summaries, appraisal drafts, landmark libraries, guideline quick-reference cards, and the Radar desk briefings) run on the operator's key behind hard, app-wide monthly ceilings, so total spend stays bounded no matter how many people use it. The whole app is a single FlaskA lightweight Python web framework. The whole app runs as a single Flask process. process over SQLiteA small, fast database that lives in a single file on disk, with no separate database server to run or maintain.: small and inspectable. As usage grows, so do the hosting and model costs, so this may move to a low monthly subscription, or annual adoption by programs and departments, in exchange for expanded capability. For now it is open.
How much of this is AI, and where the risk of error or hallucination could be. The overwhelming majority of what Foreground shows you, well over 95%, is not AI: the feed, search, evidence type, citation counts, and retraction flags are pulled straight from PubMed and public databases. The pieces that involve judgment, and can be wrong, are narrow: the relevance rating is a model's estimate of how on-topic a paper is (not a quality grade), the study-design label occasionally misreads a paper's type, the Compare tool (Beta) uses an AI pass to judge which studies actually test the same question and which way each one points (its best current method for assembling comparable evidence, but also a place where it can misjudge a study's eligibility or direction and skew the split), and the AI appraisal draft is a starting point from the abstract, never your own finished work. Two more pieces are AI-generated text and can contain errors: the guideline quick-reference cards and the Radar desk briefings are both written by a model from the underlying abstracts and item summaries, so they can misstate or omit a detail, always confirm against the guideline or the source item, which are linked. The optional AI assistant is grounded in the abstracts and open-access full text and cites its sources, but it is not infallible. For all of these, the paper is the source of truth, and the appraisal is yours.
It helps you find, frame, and organize. The appraisal is always yours.
Want the shorter version? The about page is a quick overview of what Foreground does.
Built on public data: PubMed and the NCBI E-utilities, NIH iCite, Europe PMC, OpenAlex, ClinicalTrials.gov, and openFDA. Open-access full text is aggregated by CORE (Knoth et al., CORE: A Global Aggregation Service for Open Access Papers, Nature Scientific Data 10, 366, 2023).